双层结构突触仿生忆阻器的时空信息传递及稳定性

现有计算机体系架构下的神经网络难以对多任务复杂数据进行高效处理,成为制约人工智能技术发展的瓶颈之一,而人脑的并行运算方式具有高效率、低功耗和存算一体的特点,被视为打破传统冯·诺依曼计算体系最具潜力的运算体系.突触仿生器件是指从硬件层面上实现人脑神经拟态的器件,它可以模拟脑神经对信息的处理方式,即"记忆"和"信息处理"过程在同一硬件上实现,这对于构建新的运算体系具有重要的意义.近年,制备仿生突触器件的忆阻材料已获得进展,但多聚焦于神经突触功能的模拟,对于时空信息感知和传递的关键研究较为缺乏.本文通过制备一种双层结构忆阻器,实现了突触仿生器件的基本功能包括双脉冲易化和抑制、脉冲时间依赖突触可塑性(spiking time dependent plasticity, STDP)和经验式学习等,还对器件的信息感知、传递特性和稳定性进行了研究,发现该器件脉冲测试结果满足神经网络处理时空信息的基本要求,这一结果可以为忆阻器在类脑芯片中的应用提供参考.     

基于层状多元金属氧化物的人造突触

神经形态电子学的迅速发展为生物神经系统仿生与模拟提供了有力支持.具有三明治结构的两端人造突触电子器件不仅在结构上模拟了生物突触,同时在类神经电脉冲信号的作用下可以完成对生物突触塑性的模拟与调控.本文利用溶胶-凝胶法合成了具有层状结构的P3相Na_2/3Ni_1/3Mn_2/3O₂多元金属氧化物.借助其晶体结构中Na~+易于嵌入/脱出的特性,设计并制备了基于Na_2/3Ni_1/3Mn_2/3O₂的离子迁移型人造突触,器件在电脉冲信号的刺激下实现了对生物突触塑性的模拟,并通过调校类神经尖峰脉冲信号,成功对塑性行为进行了调控.成功模拟了兴奋性突触后电流、双脉冲易化、脉冲数量依赖可塑性、脉冲频率依赖可塑性、脉冲电压幅值依赖可塑性和脉冲持续时间依赖可塑性.同时,器件实现了对摩斯电码指令的准确识别与响应.     

一种改进的WOx忆阻器模型及其突触特性分析

忆阻器被定义为第四种基本电子元器件, 其模型的研究呈现多样性. 目前, 忆阻器模型与忆阻器实际特性的切合程度引起了研究者的广泛关注. 通过改变离子扩散项, 提出了一种新的WO_x忆阻器模型, 更好地匹配了忆阻器的实际行为特性. 首先, 新的模型不仅能够描述忆阻器的一般特性, 而且能够俘获记忆丢失行为. 另外, 将新的忆阻器作为神经突触, 分析了脉冲速率依赖可塑性、短期可塑性、长期可塑性, 并发现了与生物系统中极为相似的“经验学习”现象. 最后, 考虑到温度与离子扩散系数的关系, 探讨了温度对突触权值弛豫过程的影响. 实验表明, 新忆阻器模型比原来的模型更切合实际, 且更适合作为突触而应用到神经形态系统之中.     

氧化物基忆阻型神经突触器件

忆阻器具有高密度、低功耗和阻值能够连续可调的特性,被认为是模拟神经突触最具潜力的候选者.而金属氧化物,因其氧离子可迁移,组分易于调控,与传统CMOS兼容等优点,是发展高性能忆阻器件的理想材料.本文首先介绍了氧化物基忆阻器件阻变行为及其运行机制,包括数字型和模拟型忆阻器.主要综述了基于模拟型忆阻器实现的突触器件认知功能模拟,包括非线性传输特性、时域突触可塑性、经验式学习和联合式学习等.然后进一步介绍了忆阻型突触器件在模式识别、声音定位、柔性可穿戴和光电神经突触方面的潜在应用.最后总结展望氧化物基忆阻神经突触在相关领域的可能发展方向.     

基于忆阻器的脉冲神经网络硬件加速器架构设计

脉冲神经网络(spiking neural network, SNN)作为第三代神经网络,其计算效率更高、资源开销更少,且仿生能力更强,展示出了对于语音、图像处理的优秀潜能.传统的脉冲神经网络硬件加速器通常使用加法器模拟神经元对突触权重的累加.这种设计对于硬件资源消耗较大、神经元/突触集成度不高、加速效果一般.因此,本工作开展了对拥有更高集成度、更高计算效率的脉冲神经网络推理加速器的研究.阻变式存储器(resistive random access memory, RRAM)又称忆阻器(memristor),作为一种新兴的存储技术,其阻值随电压变化而变化,可用于构建crossbar架构模拟矩阵运算,已经在被广泛应用于存算一体(processing in memory, PIM)、神经网络计算等领域.因此,本次工作基于忆阻器阵列,设计了权值存储矩阵,并结合外围电路模拟了LIF(leaky integrate and fire)神经元计算过程.之后,基于LIF神经元模型实现了脉冲神经网络硬件推理加速器设计.该加速器消耗了0.75k忆阻器,集成了24k神经元和192M突触.仿真结果显示,在5...     

忆阻突触耦合环形Hopfield神经网络动力学分析及其电路实现

提出一种新型忆阻器模型, 利用标准非线性理论分析三个忆阻特性, 并设计模拟电路。基于忆阻突触, 构建一个忆阻突触耦合环形Hopfield神经网络模型。采用分岔图、李雅普诺夫指数谱、时序图等方法, 揭示与忆阻突触密切相关的特殊动力学行为。数值仿真表明: 在忆阻突触权重的影响下, 它能够产生多种对称簇发放电模式和复杂的混沌行为。实现了该忆阻环形神经网络的模拟等效电路, 并由PSIM电路仿真验证MATLAB数值仿真的正确性。     

基于铁电材料P(VDF-TrFE)调控的多级光突触晶体管

利用有机材料PDVT-10中固有的持续光电导效应,结合铁电材料P(VDF-TrFE)提供的极化电场,通过调整铁电材料的极化强度来实现对光突触器件驰豫特性的调控。模拟了突触的短期可塑性、双脉冲易化性等基本功能,并进一步实现了多级、可调光突触。此外,持续光电导效应的驰豫现象与生物突触中Ca~(2+)的流动特性相类似,可以更好地模拟生物突触行为。研究结果为开发可调光突触提供了一个新的思路。     

铝基薄膜忆阻器作为感觉神经系统的习惯化特性

感觉神经系统可在外界刺激与生物体反应之间建立联系.感觉神经系统中的最小单位神经元可直接将外界刺激传递至中枢神经,再由中枢神经通过控制和调节生物体对外界刺激作出反应.神经突触连接了相邻神经元进行脉冲信息传递功能.习惯化是神经突触在信息传递中过滤外界无关信息时的一个基本特性,可以让感觉神经系统更快速地适应外界环境变化.忆阻器模拟神经突触功能在近年获得进展,然而针对以忆阻器为基础的具有习惯化特性的神经突触以及完整神经系统的研究相对匮乏.本文利用磁控溅射技术制备了厚度约为40 nm且含铝纳米颗粒的氮化铝薄膜忆阻器,并发现这种结构忆阻器对于重复的外界刺激有明显的习惯化行为,该行为与感觉神经系统的习惯化特性极为相似.若将这种具有习惯化的神经突触与感觉神经元串联,可形成LIF(leaky integrate-and-fire)生物模型模拟完整的神经系统行为,也为忆阻器在第三代神经网络(脉冲神经网络)中的应用提供理论参考.     

钙钛矿相界面插层对SrFeOx基忆阻器的性能提升

SrFeO_x(SFO)是一种能在SrFeO_2.5钙铁石(BM)相和SrFeO₃钙钛矿(PV)相之间发生可逆拓扑相变的材料.这种相变能显著改变电导却维持晶格框架不变,使SFO成为一种可靠的阻变材料.目前大部分SFO基忆阻器使用单层BM-SFO作为阻变功能层,这种器件一般表现出突变型阻变行为,因而其应用被局限于两态存储.对于神经形态计算等应用,单层BM-SFO忆阻器存在阻态数少、阻值波动大等问题.为解决这些问题,本研究设计出BM-SFO/PV-SFO双层忆阻器,其中PV-SFO层为富氧界面插层,可在导电细丝形成过程中提供大量氧离子并在断裂过程中回收氧离子,使导电细丝的几何尺寸(如直径)在更大范围内可调,从而获得更多、更连续且稳定的阻态,可用于模拟长时程增强和抑制等突触行为.基于该器件仿真构建了全连接神经网络(ANN),在手写体数字光学识别(ORHD)数据集进行在线训练后获得了86.3%的识别准确率,相比于单层忆阻器基ANN的准确率提升69.3%.本研究为SFO基忆阻器性能调控提供了一种新方法,并展示了它们作为人工突触器件在神...     

仿生生物感官的感存算一体化系统

生物感官集感知、存储与运算为一体的架构使其可以高效并且实时地采集和处理外界信息,这样的感存算一体化架构可为物联网时代面临的传感器数据爆炸问题提供很好的解决方案.为此,本文提出仿生生物感官的感存算一体化系统,采用不同的传感器模拟生物感受器的功能,以获取环境信息,传感器输出的模拟信号输入到模拟信号处理系统进行预处理,这样信号不需要在模拟域与数字域之间转换,可极大降低功耗和延时;预处理后的信号输入类脑运算芯片中进行分析和决策,该芯片由基于忆阻器的人工突触及人工神经元组成,通过控制突触与神经元的连接方式,可以实现不同的算法架构,如全连接脉冲神经网络、卷积脉冲神经网络以及循环脉冲神经网络等;通过运行不同的神经网络,类脑运算芯片可以实现对不同传感器信号的识别、预测以及分类等任务;更进一步,将多种仿生感觉系统的识别或预测结果结合起来,就可以实现多感官融合,这样的系统架构可以用于自动驾驶及智能机器人等复杂的场景中.     

Multilevel resistive switching and synaptic behaviors in MnO-based memristor

Exploring new synaptic electronic devices that combine computing and memory is a promising strategy that fundamentally approaches intelligent machines. In this study, the multilevel resistive switching and synaptic behaviors of a MnO-based device is studied. The device is composed of Al/MnO/Ni sandwich structure, has stable resistance switching characteristics, has continuous nonvolatile memory state, can be used as electrically programmable and erasable analog memory. The gradual conductance modulation is realized by changing the compliance current and the maximum scanning voltage. The Al/MnO/Ni devices successfully mimic the basic functions of synapses, including the paired-pulse facilitation, spike-rate-dependent plasticity, excitatory postsynaptic current, short-term plasticity, long-term plasticity, and sike-timing-dependent plasticity.     

Fabrication and investigation of ferroelectric memristors with various synaptic plasticities

In the post-Moore era, neuromorphic computing has been mainly focused on breaking the von Neumann bottlenecks. Memristors have been proposed as a key part of neuromorphic computing architectures, and can be used to emulate the synaptic plasticities of the human brain. Ferroelectric memristors represent a breakthrough for memristive devices on account of their reliable nonvolatile storage, low write/read latency and tunable conductive states. However, among the reported ferroelectric memristors, the mechanisms of resistive switching are still under debate. In addition, there needs to be more research on emulation of the brain synapses using ferroelectric memristors. Herein, Cu/PbZr_0.52Ti_0.48O₃ (PZT)/Pt ferroelectric memristors have been fabricated. The devices are able to realize the transformation from threshold switching behavior to resistive switching behavior. The synaptic plasticities, including excitatory post-synaptic current, paired-pulse facilitation, paired-pulse depression and spike time-dependent plasticity, have been mimicked by the PZT devices. Furthermore, the mechanisms of PZT devices have been investigated by first-principles calculations based on the interface barrier and conductive filament models. This work may contribute to the application of ferroelectric memristors in neuromorphic computing systems.     

Recent progress on two-dimensional neuromorphic devices and artificial neural network

Mimicking biological synapses with microelectronic devices is widely considered as the first step in hardware building artificial neuromorphic networks, which is also the basis of brain-inspired neuromorphic computing. Numerous artificial neurons and synapses making up an artificial neuromorphic network have been gained wide attention due to their powerful and efficient data processing capabilities. Recently, artificial synapses, especially memristor-type and transistor-type synapses based on multifarious two-dimensional (2D) materials have been paid much attention. The unique properties of 2D materials make devices perform well in learning ability and power efficiency when mimicking synaptic behaviors, which highlights the feasibility of 2D neuromorphic devices in constructing artificial neuromorphic networks. Herein, the basic structures and principles of biological synapses are introduced, and the definitions of synaptic behaviors in synaptic electronic devices are discussed. Then, the progress of 2D memristor-type and transistor-type neuromorphic devices involving their device architecture, neuromorphic operational mechanism, and promising applications is reviewed. Finally, the future challenges of artificial synaptic devices based on 2D materials are discussed briefly.     

Electronic synapses based on ultrathin quasi-two-dimensional gallium oxide memristor

Synapse emulation is very important for realizing neuromorphic computing, which could overcome the energy and throughput limitations of today's computing architectures. Memristors have been extensively studied for using in nonvolatile memory storage and neuromorphic computing. In this paper, we report the fabrication of vertical sandwiched memristor device using ultrathin quasi-two-dimensional gallium oxide produced by squeegee method. The as-fabricated two-terminal memristor device exhibited the essential functions of biological synapses, such as depression and potentiation of synaptic weight, transition from short time memory to long time memory, spike-timing-dependent plasticity, and spike-rate-dependent plasticity. The synaptic weight of the memristor could be tuned by the applied voltage pulse, number,width, and frequency. We believe that the injection of the top Ag cations should play a significant role for the memristor phenomenon. The ultrathin of medium layer represents an advance to integration in vertical direction for future applications and our results provide an alternative way to fabricate synaptic devices.     

玉米淀粉固态电解质质子\电子杂化突触晶体管

随绿色可持续发展观念的深入人心,研究人员致力于寻找天然有机材料应用于功能性电子器件.淀粉以其低廉的价格、丰富的来源和优异的机械性能进入了科研人员的视野.淀粉可由玉米、马铃薯、甘薯和葛根等含淀粉的物质中提取而得,一般不溶于水,在和水加热至一定温度时,则糊化成胶状溶液.本文通过旋涂法将玉米淀粉的胶状溶液旋涂至氧化铟锡玻璃表面,然后在30?C恒温环境中晾干制备成固态胶合状薄膜.以此薄膜作为固态电解质制备了氧化铟锌突触晶体管,并实现了生物神经突触的双脉冲易化、学习记忆能力、高通滤波等可塑性行为的仿真.本研究以玉米淀粉固态胶合薄膜作为电解质大大降低了氧化物薄膜晶体管固态电解质的成本,且该电解质无毒性、来源丰富,将为人工神经网络的开发提供一种可选择的元件.     

基于磁电耦合效应的基本电路元件和非易失性存储器

磁电耦合效应是指磁场控制电极化或者电场控制磁性的物理现象,它们为开发新型电子器件提供了额外的物理状态自由度,具有巨大的应用潜力.磁电耦合系数作为磁电耦合材料的重要参量,体现了材料磁化和电极化的耦合性能,其随外加物理场的变化可以表现出非线性回滞行为,具备作为非易失存储的物理状态特征.本文讨论了基于磁电耦合效应如何建立起电荷-磁通之间的直接关联,继而实现了第四种基本电路元件并构建了完整的电路元件关系图.在此基础上,研究了多铁性异质结中的非线性磁电耦合效应,并利用其独特的电荷-磁通关联特性,开发了基于磁电耦合系数的电写-磁读型非易失性信息存储、逻辑计算与类神经突触记忆等一系列新型信息功能器件.     

Brain-like synaptic memristor based on lithium-doped silicate for neuromorphic computing

Artificial synapse is one of the potential electronics for constructing neural network hardware. In this work, Pt/LiSiO_x/TiN analog artificial synapse memristor is designed and investigated. With the increase of compliance current (C. C.) under 0.6 mA, 1 mA, and 3 mA, the current in the high resistance state (HRS) presents an increasing variation, which indicates lithium ions participates in the operation process for Pt/LiSiO_x/TiN memristor. Moreover, depending on the movement of lithium ions in the functional layer, the memristor illustrates excellent conduction modulation property, so the long-term potentiation (LTP) or depression (LTD) and paired-pulse facilitation (PPF) synaptic functions are successfully achieved. The neural network simulation for pattern recognition is proposed with the recognition accuracy of 91.4%. These findings suggest the potential application of the LiSiO_x memristor in the neuromorphic computing.     

基于Au/TiO_2/FTO结构忆阻器的开关特性与机理研究

采用简单的一步水热法在FTO导电玻璃上外延生长了锐钛矿TiO_2纳米线,制备了具有Au/TiO_2/FTO器件结构的锐钛矿TiO_2纳米线忆阻器,系统研究了器件的阻变开关特性和开关机理.结果表明,Au/TiO_2/FTO忆阻器具有非易失的双极性阻变开关特性.同时,在103s的时间内,器件在0.1 V的电阻开关比始终保持在20以上,表明器件具有良好的非易失性.此外,器件在低阻态时遵循欧姆导电特性,而在高阻态时则满足陷阱控制的空间电荷限制电流传导机制,同时提出了基于氧空位导电细丝形成与断开机制的阻变开关模型.研究结果表明Au/TiO_2/FTO忆阻器将是一种很有发展潜力的下一代非易失性存储器.     

Modulating 3D memristor synapse by analog spiking pulses for bioinspired neuromorphic computing

Memristor based artificial synapses have demonstrated great potential for bioinspired neuromorphic computing in recent years. To emulate synaptic functions, such as short-term plasticity and long-term potentiation/depression, square pulses or combined complex pulse groups are applied on the device. However, in biological neuron systems, the action potentials are analog pulses with similar amplitudes. Furthermore, in biological systems, the intensity of the stimulus is coded into the frequency of action potentials to modulate the weight of synapses. Toward this programming method, we applied a series of analog spiking pulses with same peaks on Ru/TiO _x /TiN 3D memristor to emulate synaptic functions, such as long-term potentiation/depression and synaptic saturation. Moreover, we demonstrated the conductance change of the device under different stimulus frequencies of analog spiking pulses and described the statistical results of conductance change value, which shows that the device conductance has a larger change value under a higher spiking frequency with identical pulse number. These results show that the analog spiking pulses can well modulate the memristor-based synaptic weight and have a great potential for bioinspired computing in the future.